There’s been a sense of increasing unease from some professional bodies and courts about the extent to which AI might already be starting to take the place of lawyers’ decision making functions and how so-called “Agentic AI”[1] might serve to accelerate that process. Helen Evans KC and Marie-Claire O’Kane examine what’s been going on- and the likely next steps.
In its recent paper “Shaping the Future of Agentic AI in Legal Practice” the Law Society identified that as a result of AI, many law firms are already moving over to business models that require less direct human supervision. For instance, lawyers using AI for large scale tasks such as document reviews are increasingly adopting a sampling approach to oversight, rather than a more hands-on approach. But matters are likely to go much further in future because, as the Law Society puts it, “unlike generative AI, which produces content within guardrails of human initiation, agentic AI can pursue objectives over time, across platforms, and with minimal supervision.”
The Law Society makes clear that it already regards an approach based on “validating whether the system behaves reliably” as a significant departure from “existing professional assumptions about responsibility and control”- so what might the future hold if Agentic AI or other methods of AI working “independently” do come increasingly to the fore?
What the courts have had to say about AI and oversight so far
The courts have now considered the role of AI on multiple occasions, although largely focusing to date on disclosure and hallucinations. In all of the core cases so far they have emphasised the importance not only of keeping humans in the loop- but also that the humans using AI are appropriately supervised and operate within ethical frameworks.
Although the cases to date have all concerned the conduct of litigation rather than other aspects of lawyers’ work, there is a question about how the approach taken would tally with AI increasingly taking over workstreams, or other practices developing where AI is used with no input from lawyers.
Disclosure
In Gormsen v Meta Platforms, Inc. and Others [2025] CAT 85 the Competition Appeal Tribunal considered the parties’ use of technology in a large scale disclosure task which it described at [202] as “an important, massive and complex exercise”. The Tribunal was willing to facilitate the use of TAR tools or AI, without being prescriptive about what tool had to be used. However, despite acknowledging the potential for technology to cut through large document sets, it firmly kept the lawyers in the driving seat:
- The Tribunal proceeded to set out detailed and prescriptive requirements for the lawyers in the disclosure team, including the need for “guidance, an audit trail and quality control”, an “induction pack” for team members and a “disclosure log” [208-210];
- The Tribunal noted that [240] that the Designated Solicitor for each party should “monitor and review” the use of technology and be willing to adapt or modify the approach;
- At [241] the Tribunal took particular comfort from the fact that “there appear to be adequate procedures imposing stringent human tests and checks on the approach.” It determined at [242] that it did not consider the use of AI technology in the document review to be “experimental” in circumstances where it is used “to supplement and assist human review.”
Hallucinations
The hallucination cases have taken a similar tack.
- Starting with R (Ayinde) v LB Haringey, Al-Haroun v Qatar National Bank [2025] EWHC 1383, the Divisional Court made clear that AI use had to be used with an “appropriate degree of oversight, and within a regulatory framework that ensures compliance with well-established professional and ethical standards if public confidence in the administration of justice is to be maintained” [5];
- In UK and R (on the application of Munir) v Secretary of State for the Home Department [2026] UKUT 81 the Upper Tribunal at [37] emphasised that the qualified legal professional with conduct of a matter is expected to ensure that documents are checked, that errors are identified, and that only accurate documents are sent to the Tribunal;
- More recently still, in Cork and another v Mark Smith [2026] EWHC 1199 (Ch) the court stated at [95] that: “legal professionals bear ultimate responsibility for their work and cannot outsource the process of legal research or of legal reasoning to an AI. It is a tool to be used with caution. AI has the potential to be wholly unreliable. AI may of course provide a jumping off point for research and legal reasoning but it does not, at least at present, do away with the need for proper research and thought on the part of a legal professional, even a very junior legal professional.”
What business models are emerging?
In March 2025, the SRA authorised Garfield Law to provide an AI service handling repetitious and formulaic tasks in debt recovery litigation. However the authorisation included important brakes on the service. Garfield is not permitted to suggest case law- which is a known problem area for hallucinations. Furthermore, it is not autonomous. A step can only be taken in the debt recovery litigation once a client has approved it. Finally, named regulated solicitors are responsible for the work.
Alongside this, unregulated competitors have entered the market.[2] One such service describes itself as an “AI Law firm – not a traditional law firm” and a “fully automated law firm with no lawyers doing the cases”. Its website goes on to say that it is “not a regulated entity; this makes it fast efficient and available to everyone.”
These examples may be extreme in the sense that they are AI only services. However it seems to us likely that some regulated law firms are likely to seek to develop AI only offerings as part of their overall services. Examples might be an “AI-only contract drafting” service.
The courts have in the past proven willing to support other innovative and costs saving business models. A prime example would be “unbundled retainers” – whereby a professional only deals with a discrete task such as drawing up but not advising on the terms of an agreement: see Minkin v Landsberg [2016] 1 WLR 1489 . In that context, the Court of Appeal upheld a retainer that was considerably narrower than a more “traditional” service.
However, to our knowledge, no case has yet considered whether it is appropriate for lawyers to enter into agreements with their clients where part of the work is effectively delegated to AI or handled with only minimal human input. If the observations in the disclosure and hallucination cases to date apply more broadly, the courts might take some convincing that this was appropriate. This is particularly the case where the work touches on litigation, where the court has to be able to depend on law firms performing their duties not only to their clients but to the court itself.
We note that the limited array of AI cases to date have all focused on that context and have not yet considered the place of AI in advisory or transactional work. Whether the courts and regulators are willing to take a more relaxed view towards questions of supervision and accountability in that context remains to be seen.
Where next?
It is not easy to identify the pace of change on the ground. On the one hand, the Law Society’s work on Agentic AI suggests that it anticipates an incremental drift into lawyers using AI in a way that substitutes machines for professional judgment. However, it also expresses concern that pioneering practices are already emerging faster than the regulators can keep up:
“Despite strong professional norms requiring human sign-off on legal work, many legal teams are already experimenting with governance practices that anticipate more autonomous systems.”
The Law Society’s work on Agentic AI suggests that it is “preparing for disruption”. Lawyers can be forgiven for feeling that generative AI has already made enough waves; however the Law Society suggests that Agentic AI may come into even starker conflict with the traditional approach to regulating the legal profession.
© Helen Evans KC and Marie-Claire O’Kane of 4 New Square Chambers
June 2026
This article is not intended as a substitute for legal advice. Advice about a given set of facts should always be taken.
See 4 New Square’s AI Hub: www.4newsquare.com/ai-hub/
[1] Of course, as the UK Joint Taskforce’s Consultation Paper on Liability for AI Harms points out- AI does not in fact have legal personality and the Paper views it as a “tool”.
[2] Unregulated law firms have to ensure that they do not start taking on reserved activities within the meaning of the Legal Services Act 2007.
